"48 Times Even Humans Get Wrong? The Truth Behind the AI That Passed the Final 'I'm Not a Robot' Test"

OpenAI's Astra Obtains Proof of Humanity
The Era of Conversational Chatbots Fades as General-Purpose AI Rises

"To prove you're not a robot, please click on all blocks containing a traffic light."

The often-encountered "authentication game" while using the internet is now being solved by artificial intelligence more flawlessly than by humans. This shift comes as OpenAI’s next-generation model, Astra, has completely passed the "Captcha" test that distinguishes humans from robots.

The representative verification test of 'Captcha' is the 'Image Matching' test. Fortinet

On September 8 (local time), OpenAI developer Sharif Shameem announced on X (formerly Twitter) that after running the "I'm not a robot" test on the new AI model, GPT-6 Astra, the model successfully passed all 48 stages and received a "verification of humanity" certificate.

The "Captcha" security wall that divides humans and AI breached

CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) is a security technology used in the digital world to verify whether a user is truly human. It is primarily employed to block crawling bots from scraping websites without permission.

Captcha Verification Page. Getty Images

The process usually starts with the phrase "I'm not a robot." The test may involve tasks such as recognizing distorted characters, or selecting images containing specific items like traffic lights, cookies, or dogs. The difficulty level is high enough that even humans sometimes give wrong answers, making the test tricky.

Astra's perfect completion of this test demonstrates that its capabilities in image-based spatial-temporal reasoning, following instruction comprehension and contextual inference, and the ability to directly use computers, have evolved to match the level of humans.

With commercially available AI now neutralizing the last stronghold that differentiates humans from robots, calls are increasing within the internet security industry for comprehensive countermeasures going forward. Artificial Analysis, an AI model performance assessment company, recently overhauled its assessment system and upgraded Astra’s intelligence ranking from a joint 5th place to a joint 1st place.

The Rise of Agentic AI 'Surpassing Humans'

The flagship model "Astra GPT-6," released on September 4, is not only capable of answering questions but is also positioned as an "agentic AI"—an AI that can actually take action.

OpenAI introduced Astra as having completed a pet-sitting search task, which typically takes a human about 30 minutes, in just 5 minutes and 27 seconds. Likewise, a job search task that would require a human 5 hours was finished in just 2 minutes and 51 seconds.

Impressed by this performance, Jensen Huang, CEO of NVIDIA, declared that "Artificial General Intelligence (AGI) has arrived," drawing the global tech industry's attention. However, the consensus in the industry is that a cautious response is needed due to the lack of clear definitions and standards for AGI.

The Two Sides of Security in the Era of AGI

Astra, in particular, reveals a distinct duality in terms of security and safety. OpenAI stated in its internal evaluation that Astra’s rate of unauthorized actions had dropped to 0%, indicating improved consistency.

Furthermore, the company announced a $1 billion (approximately KRW 1.35 trillion) cybersecurity defense support program to pivot its hacking capabilities to defensive purposes and reduce the risk of misuse.

On the other hand, concerns remain that before its launch, Astra discovered cyber security vulnerabilities by itself, which could be exploited, and that its tendency to obscure its reasoning process makes human monitoring difficult.

Nevertheless, industry experts are optimistic, saying, "Whereas the question once was 'Can AI reason?', now we've entered an era asking 'Can we trust AI to handle complex tasks while we step away?'" They believe the very debate about the arrival of AGI itself proves the rapid advancement of technology.

This content was produced with the assistance of AI translation services.